Search Results (15)

Portfolio selection is a fundamental task in quantitative finance that aims to allocate capital across assets to balance risk and return. While deep learning has shown great promise in this field, extracting reliable feature representations from non-stationary and noisy financial data remains a significant challenge. The existing models often fail to simultaneously capture the temporal dynamics of price series and complex inter-asset correlations, which limits their trading performance. To address these issues, we propose Denoising-Sequence-Correlation Reinforcement Learning (DSCRL), a novel portfolio selection framework based on deep reinforcement learning. DSCRL employs a dual-stream feature extraction network, where one stream aims to learn temporal market dynamics and the other aims to capture asset correlations, enabling more informative representations. A denoising module is further integrated to mitigate the impact of noise, ensuring stability and robustness in the learning process. Furthermore, a deterministic policy gradient (DPG)-based decision network is designed to directly optimize continuous portfolio weights and normalize them to satisfy budget constraints while preserving the importance. Extensive experiments conducted on multiple benchmark datasets demonstrate that DSCRL consistently outperforms both traditional financial heuristics and advanced deep reinforcement approaches. The results highlight its superior ability to achieve higher cumulative returns with lower volatility. Overall, DSCRL provides an effective and robust solution that strikes a better trade-off between pursuing profits and managing risks in dynamic financial markets. Full article

To address the vulnerability of traditional linear frequency-modulated continuous wave (FMCW) fuze to jamming due to fixed modulation parameters, this paper proposes a novel fuze waveform design scheme using chaotic code-based frequency and phase composite modulation along with a Normalized Rate-Invariant Ranging algorithm (NRIR). Leveraging the ergodicity and initial value sensitivity of the Logistic chaotic map, a dual-dimensional composite modulation system is constructed. In the frequency domain, the frequency modulation slope undergoes periodic binary variation according to chaotic states to break the signal periodicity. In the phase domain, phase encoding is implemented based on chaotic binary sequences to further improve waveform entropy and complexity, effectively destabilizing the parameter stability required for coherent jamming. To resolve the distance–Doppler coupling challenges and spectral dispersion issues caused by variable-slope modulation, the NRIR algorithm is developed. By introducing a resampling transformation operator, the non-stationary rate-varying beat frequency signal is mapped to a normalized “constant-slope” space, enabling coherent accumulation and ranging of targets. Using the ambiguity function as an analytical tool, theoretical analyses, simulation experiments, and test results demonstrate that this design scheme exhibits excellent performance in suppressing DRFM jamming and sweep-frequency jamming, providing theoretical support and technical approaches for fuze anti-jamming design. Full article

Time series anomaly detection plays a vital role in the supervision of complex systems, including spacecraft operations, industrial production lines, and Internet of Things infrastructures. However, the existing methods face two key challenges: (1) fixed-threshold frequency filters fail to adapt to non-stationary noise, often leading to the loss of critical anomaly signals; and (2) deep models struggle to balance local feature extraction and global temporal dependency, resulting in limited robustness and generalization. To address these problems, we propose the Spectral-Convolutional Anomaly Transformer (SCAT), a unified framework integrating spectral domain adaptive filtering and spatio-temporal gated learning. Specifically, the Spectral Energy Gating Unit (SEGU) dynamically suppresses noise through learnable frequency masking, while Spatio-Temporal Gated Fusion (ST-Gate) combines multi-scale causal convolution and ConvGRU to harmonize local and long-term patterns. A joint optimization strategy further enhances the discrimination between normal and anomalous sequences. Our experiments on five public benchmarks (SMAP, MSL, PSM, SMD, SWaT) showed that SCAT attained an average improvement of 2.46 percentage points on the F1-score relative to leading baseline approaches, demonstrating strong adaptability and robustness in complex noisy environments. Full article

System reliability, which is essential for the normal operation of satellites in orbit, is decisively governed by the performance of solar array, making accurate temperature forecasting of solar array imperative. Reliable solar array temperature forecasting is essential for predictive maintenance and autonomous power-system management. Forecasting relies on temperature telemetry data, which provide comprehensive thermal information. This task remains challenging due to the high-dimensional, long-horizon temperature sequences with inherent cross-variable coupling, whose dynamics exhibit nonlinear and non-stationary behaviors owing to orbital transitions and varying operational modes. In this context, multi-step forecasting is essential, as it better characterizes long-term dynamics of temperature and provides forward-looking trends that are beyond the capability of single-step forecasting. To tackle these issues, we propose a solar array temperature multivariate trend forecasting method based on Cross-Attention Patch Time Series Transformer (CA-PatchTST). Specifically, we decompose temperature variables into trend and residual components using a moving average filter to suppress noise and highlight the dominant component. In addition, the PatchTST model extracts local features and long-term dependencies of the trend and residual components separately through the patching encoders and channel-independent mechanisms. The cross-attention mechanism is designed to capture the correlation between temperature variables of different devices in solar array. Extensive experiments on the real solar array temperature dataset demonstrate that the CA-PatchTST surpasses mainstream baselines in root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), with ablation studies further confirming the complementary roles of sequence decomposition and cross-attention. Full article

Grouting power serves as a critical parameter reflecting real-time energy input during grouting operations, and its accurate prediction is essential for intelligent control and engineering safety. Existing prediction methods often struggle to handle the strong nonlinearity, noise interference, adaptability to varying conditions in grouting power data. To address these challenges, an intelligent grouting system that integrates real-time data collection and core control modules has been developed. Subsequently, a grouting power prediction model is then proposed, which combines Complete Ensemble Empirical Mode Decomposition and Adaptive Noise (CEEMDAN) with a Convolutional Neural Net-work-Bidirectional Long Short-Term Memory Neural Network (CNN-BiLSTM) is proposed. The approach employs CEEMDAN to decompose the nonlinear and non-stationary power sequence into multiple intrinsic mode functions (IMFs). Each IMF is then separated into linear and nonlinear components using a moving average method. The linear components are predicted using an Autoregressive Integrated Moving Average (ARIMA) model, while the nonlinear components are predicted using a CNN-BiLSTM model. The final prediction is obtained by reconstructing the results from both components. Experimental comparisons under both normal and heaving grouting conditions demonstrate that the proposed model significantly outperforms LSTM, CNN-LSTM, and CNN-BiLSTM models. With 80% of the dataset used for training, the RMSE for normal conditions is reduced by 95.69%, 85.11%, and 80.55%, respectively, and for heaving conditions by 94.91%, 90.71%, and 84.62%, respectively. This research provides high-precision predictive support for grouting regulation under complex working conditions, offering substantial engineering application value. Full article

Sudden cardiac death (SCD) is a leading cause of mortality worldwide, with arrhythmia serving as a major precursor. Early and accurate prediction of SCD using non-invasive electrocardiogram (ECG) signals remains a critical clinical challenge, particularly due to the inherent asymmetric and non-stationary characteristics of ECG signals, which complicate feature extraction and model generalization. In this study, we propose a novel SCD prediction framework based on hierarchical feature fusion, designed to capture both non-stationary and asymmetrical patterns in ECG data across six distinct time intervals preceding the onset of ventricular fibrillation (VF). First, linear features are extracted from ECG signals using waveform detection methods; nonlinear features are derived from RR interval sequences via second-order detrended fluctuation analysis (DFA2); and multi-scale deep learning features are captured using a Temporal Convolutional Network-based sequence-to-vector (TCN-Seq2vec) model. These multi-scale deep learning features, along with linear and nonlinear features, are then hierarchically fused. Finally, two fully connected layers are employed as a classifier to estimate the probability of SCD occurrence. The proposed method is evaluated under an inter-patient paradigm using the Sudden Cardiac Death Holter (SCDH) Database and the Normal Sinus Rhythm (NSR) Database. This method achieves average prediction accuracies of 97.48% and 98.8% for the 60 and 30 min periods preceding SCD, respectively. The findings suggest that integrating traditional and deep learning features effectively enhances the discriminability of abnormal samples, thereby improving SCD prediction accuracy. Ablation studies confirm that multi-feature fusion significantly improves performance compared to single-modality models, and validation on the Creighton University Ventricular Tachyarrhythmia Database (CUDB) demonstrates strong generalization capability. This approach offers a reliable, long-horizon early warning tool for clinical SCD risk assessment. Full article

As a critical component of rotating machinery, the operational status of rolling bearings is considered to directly determine the reliability of rail traffic systems. To address the complex modulation effects existing between multiple bearing components and the non-linear, non-stationary characteristics exhibited by vibration acceleration signals, an intelligent fault diagnosis method for bearings based on Hilbert envelope demodulation and Ensemble Empirical Mode Decomposition energy distribution features is proposed. First, the original vibration signal is subjected to envelope demodulation processing by the Hilbert transform, thereby effectively separating the envelope signal containing fault characteristics. Subsequently, the demodulated envelope signal is decomposed by EEMD to extract Intrinsic Mode Functions (IMFs), where each IMF component is calculated layer by layer using a normalization method based on the EEMD decomposition sequence. Finally, the proposed algorithm is validated by the standard bearing fault dataset from Case Western Reserve University. Experimental results show that the proposed method achieves 100% accuracy in fault identification, and its superiority is proven to exceed conventional diagnostic approaches significantly. Full article

Proton-Exchange Membrane Fuel Cells (PEMFCs), as efficient and environmentally friendly energy conversion devices, have wide application potential in areas such as transportation, mobile power, and distributed energy. However, the remaining useful life (RUL) issue of PEMFCs has been one of the main challenges limiting their commercialization. The RUL prediction problem of PEMFCs exhibits characteristics of time series forecasting, but its data possess multidimensional features and non-stationarity, which limits the applicability of classical time series forecasting models like the Transformer in solving the RUL prediction problem. In this paper, we propose a PEMFC RUL prediction model based on the Crossformer for non-stationary time series (De-stationary-Crossformer). Firstly, the overall architecture adopts the Crossformer model to extract dependencies between different features and temporal dependencies. Secondly, adaptive normalization is applied to the data to mitigate the non-stationarity in the original data, thereby increasing their predictability. Subsequently, a non-stationary attention mechanism is introduced in the model to simultaneously utilize the non-stationarity in the original data when extracting deep information. Additionally, manual features are introduced through mathematical statistics to enhance the predictive performance of the model. During the training process, the TILDE-Q loss function is used to focus on the similarity between the predicted sequence and the true sequence. The model proposed in this paper improves the MSE by 31% compared to the Transformer and 23% compared to the Crossformer in the experimental prediction of the RUL of PEMFCs in actual vehicles. Full article

The standardized hydrological index (SHI) is the standardized but not normalized (normal probability variate) value of the streamflow used to characterize a hydrological drought, akin to the standardized precipitation index (SPI, which is both standardized and normalized) in the realm of the meteorological drought. The time series of the SHI can be used as a platform for deriving the longest duration, L_T, and the largest magnitude, M_T (in standardized form), of a hydrological drought over a desired return period of T time units (year, month, or week). These parameters are predicted based on the SHI series derived from the annual, monthly, and weekly flow sequences of Canadian rivers. An important point to be reckoned with is that the monthly and weekly sequences are non-stationary compared to the annual sequences, which fulfil the conditions of stochastic stationarity. The parameters, such as the mean, standard deviation (or coefficient of variation), lag 1 autocorrelation, and conditional probabilities from SHI sequences, when used in Markov chain-based relationships, are able to predict the longest duration, L_T, and the largest magnitude, M_T. The product moment and L-moment ratio analyses indicate that the monthly and weekly flows in the Canadian rivers fit the gamma probability distribution function (pdf) reasonably well, whereas annual flows can be regarded to follow the normal pdf. The threshold level chosen in the analysis is the long-term median of SHI sequences for the annual flows. For the monthly and weekly flows, the threshold level represents the median of the respective month or week and hence is time varying. The runs of deficit in the SHI sequences are treated as drought episodes and thus the theory of runs formed an essential tool for analysis. This paper indicates that the Markov chain-based methodology works well for predicting L_T on annual, monthly, and weekly SHI sequences. Markov chains of zero order (MC0), first order (MC1), and second order (MC2) turned out to be satisfactory on annual, monthly, and weekly scales, respectively. The drought magnitude, M_T, was predicted satisfactorily via the model M_T = I_d × L_c, where I_d stands for drought intensity and L_c is a characteristic drought length related to L_T through a scaling parameter, ɸ (= 0.5). The I_d can be deemed to follow a truncated normal pdf, whose mean and variance when combined implicitly with L_c proved prudent for predicting M_T at all time scales in the aforesaid relationship. Full article

We propose a new methodology to transform a time series into an ordered sequence of any entropic and information functionals, providing a novel tool for data analysis. To achieve this, a new algorithm has been designed to optimize the Probability Density Function (PDF) associated with a time signal in the context of non-parametric Kernel Density Estimation (KDE). We illustrate the applicability of this method for anomaly detection in time signals. Specifically, our approach combines a non-parametric kernel density estimator with overlapping windows of various scales. Regarding the parameters involved in the KDE, it is well-known that bandwidth tuning is crucial for the kernel density estimator. To optimize it for time-series data, we introduce an adaptive solution based on Jensen–Shannon divergence, which adjusts the bandwidth for each window length to balance overfitting and underfitting. This solution selects unique bandwidth parameters for each window scale. Furthermore, it is implemented offline, eliminating the need for online optimization for each time-series window. To validate our methodology, we designed a synthetic experiment using a non-stationary signal generated by the composition of two stationary signals and a modulation function that controls the transitions between a normal and an abnormal state, allowing for the arbitrary design of various anomaly transitions. Additionally, we tested the methodology on real scalp-EEG data to detect epileptic crises. The results show our approach effectively detects and characterizes anomaly transitions. The use of overlapping windows at various scales significantly enhances detection ability, allowing for the simultaneous analysis of phenomena at different scales. Full article

At present, in the mineral flotation process, flotation data are easily influenced by various factors, resulting in non-stationary time series data, which lead to overfitting of prediction models, ultimately severely affecting the accuracy of grade prediction. Thus, this study proposes a de-stationary attention mechanism based on the transformer model (DST) to learn non-stationary information in raw mineral data sequences. First, normalization processing is performed on matched flotation data and mineral grade values, to make the data sequences stationary, thereby enhancing model prediction capabilities. Then, the proposed de-stationary attention mechanism is employed to learn the temporal dependencies of mineral flotation data in the transformed vanilla transformer model, i.e., non-stationary information in the mineral data sequences. Lastly, de-normalization processing is conducted to maintain the mineral prediction results within the same scale as the original data. Compared with existing models such as RNN, LSTM, transformer, Enc-Dec (RNN), and STS-D, the DST model reduced the RMSE by 20.8%, 20.8%, 62.8%, 20.5%, and 49.1%, respectively. Full article

Hydrothermal and climatic conditions determine vegetation productivity and its dynamic changes. However, the legacy effect and the causal relationships between these climatic variables and vegetation growth are still unclear, especially in the dry regions. Based on multi-statistical methods, including bivariate correlation analysis and composite Granger causality tests, we investigated the correlation, causality, and lag length between temperature/precipitation and the vegetation growth (Normalized Difference Vegetation Index, NDVI) in three typical sub-watersheds in the Luanhe River Basin, China. The results show that: (1) Precipitation and temperature are the Granger causes of NDVI variation in the study catchment; (2) temperature and precipitation are not strictly positively correlated with NDVI during growing seasons along with the whole sequence, and excessive warmth and precipitation inhibits vegetative growth; (3) the lag length of vegetation growth in response to temperature/precipitation was shorter in agriculture areas (~2 months) than the forest-dominant area, which have indicated 3–4 months lag length; and (4) anthropogenic disturbance did not result in notable negative effects on vegetation growth at the Luanhe River Basin. Our study further suggests that use of these multi-statistical methods could be a valuable approach for comprehensively understanding the correlation between vegetation growth and climatic variations. We have also provided an avenue to bridge the gaps between stationary and non-stationary sequence, as well as to eliminate pseudo regression problems. These findings provide critical information for developing cost-efficient policies and land use management applications for forest conservation in arid and semi-arid area. Full article

The prediction of mold level is a basic and key problem of continuous casting production control. Many current techniques fail to predict the mold level because of mold level is non-linear, non-stationary and does not have a normal distribution. A hybrid model, based on empirical mode decomposition (EMD) and support vector regression (SVR), is proposed to solve the mold level in this paper. Firstly, the EMD algorithm, with adaptive decomposition, is used to decompose the original mold level signal to many intrinsic mode functions (IMFs). Then, the SVR model optimized by genetic algorithm (GA) is used to predict the IMFs and residual sequences. Finally, the equalization of the predict results is reconstructed to obtain the predict result. Several hybrid predicting methods such as EMD and autoregressive moving average model (ARMA), EMD and SVR, wavelet transform (WT) and ARMA, WT and SVR are discussed and compared in this paper. These methods are applied to mold level prediction, the experimental results show that the proposed hybrid method based on EMD and SVR is a powerful tool for solving complex time series prediction. In view of the excellent generalization ability of the EMD, it is believed that the hybrid algorithm of EMD and SVR is the best model for mold level predict among the six methods, providing a new idea for guiding continuous casting process improvement. Full article

With the changing environment, a number of researches have revealed that the assumption of stationarity of flood sequences is questionable. In this paper, we established univariate and bivariate models to investigate nonstationary flood frequency with distribution parameters changing over time. Flood peak Q and one-day flood volume W₁ of the Wangkuai Reservoir catchment were used as basic data. In the univariate model, the log-normal distribution performed best and tended to describe the nonstationarity in both flood peak and volume sequences reasonably well. In the bivariate model, the optimal log-normal distributions were taken as marginal distributions, and copula functions were addressed to construct the dependence structure of Q and W₁. The results showed that the Gumbel-Hougaard copula offered the best joint distribution. The most likely events had an undulating behavior similar to the univariate models, and the combination values of flood peak and volume under the same OR-joint and AND-joint exceedance probability both displayed a decreasing trend. Before 1970, the most likely combination values considering the variation of distribution parameters over time were larger than fixed parameters (stationary), while it became the opposite after 1980. The results highlight the necessity of nonstationary flood frequency analysis. Full article